Modern quantum chemical methods involve a trade-off between accuracy and computational cost/complexity. As an alternative, deep learning methods are used as shortcuts to create accurate predictions with small computational complexity. Such models proved to be highly effective in predicting closed-shell systems (where all electrons are paired). However, little focus has been given to predicting open-shell systems (where there are unpaired electrons), despite their importance in describing species like radicals and reaction intermediates. We introduce OrbNet-Spin, built upon OrbNet-Equi, a geometric- and quantum-aware deep learning model for representing chemical systems at the electronic structure level. OrbNet-Spin incorporates a spin-polarized treatment into the underlying semiempirical quantum mechanics orbital featurization and adjusts the model architecture accordingly while maintaining the geometrical constraints. OrbNet-Spin can accurately describe both closed and open-shell electronic structures. We validate OrbNet-Spin's performance using the QMSpin dataset of open-shell carbenes, achieving a mean absolute error below the chemical accuracy in both singlet and triplet carbenes.